fix(trainer): 修复检查点加载逻辑

khaosz/config/param_config.py

@@ -119,6 +119,14 @@ class Checkpoint(BaseModelIO):
         default_factory=list,
         metadata={"help": "List of training losses."}
     )
+    epoch: int = field(
+        default=0,
+        metadata={"help": "Current epoch."}
+    )
+    batch_iter: int = field(
+        default=0,
+        metadata={"help": "Current iteration."}
+    )
     
     def _get_training_paths(self, directory: Union[str, Path]) -> dict[str, Path]:
         paths = self._get_file_paths(directory)
@@ -173,11 +181,11 @@ class Checkpoint(BaseModelIO):
         if not self.loss_list:
             return
         
-        current_iter = len(self.loss_list)
+        batch_iter = len(self.loss_list)
         
         plt.figure(figsize=(10, 6))
         plt.plot(self.loss_list)
-        plt.title(f"Training Loss - Iteration {current_iter}")
+        plt.title(f"Training Loss - Iteration {batch_iter}")
         plt.xlabel("Batch")
         plt.ylabel("Loss")
         plt.grid(True)
@@ -234,5 +242,3 @@ class ParameterLoader:
             loss_list=loss_list or [],
             optimizer_state=optimizer
         )
-
-

khaosz/data/data_util.py

@@ -274,13 +274,14 @@ class ResumeableRandomSampler(Sampler[int]):
         generator = torch.Generator()
         generator.manual_seed(seed)
 
+        # consume  previous epochs
+        for _ in range(start_epoch):
+            torch.randperm(self.num_samples, generator=generator)
+        
         self.generator = generator
         self._indices = None
     
     def _get_indices(self):
-        for _ in range(self.epoch):
-            _ = torch.randperm(self.num_samples, generator=self.generator)
-        
         current_epoch_indices = torch.randperm(self.num_samples, generator=self.generator).tolist()
         self._indices = current_epoch_indices[self.iter % self.num_samples:]
     

khaosz/trainer/train_callback.py

@@ -96,20 +96,22 @@ class CheckpointCallback(TrainCallback):
         self.last_ckpt_iter = 0
     
     def _save_checkpoint(self, trainer: 'Trainer', context: 'TrainContext'):
-        save_path = os.path.join(trainer.train_config.checkpoint_dir, f"iter_{context.current_iter}")
+        save_path = os.path.join(trainer.train_config.checkpoint_dir, f"iter_{context.batch_iter}")
         context.checkpoint.optimizer_state = context.optimizer.state_dict()
         context.checkpoint.scheduler_state = context.scheduler.state_dict()
+        context.checkpoint.epoch = context.epoch
+        context.checkpoint.batch_iter = context.batch_iter
         context.checkpoint.save(save_path)
-        self.last_ckpt_iter = context.current_iter
+        self.last_ckpt_iter = context.batch_iter
     
     def on_batch_end(self, trainer: 'Trainer', context: 'TrainContext'):
         context.checkpoint.loss_list.append(context.loss)
         
-        if context.current_iter - self.last_ckpt_iter >= self.checkpoint_interval:
+        if context.batch_iter - self.last_ckpt_iter >= self.checkpoint_interval:
             self._save_checkpoint(trainer, context)
             
     def on_train_end(self, trainer: 'Trainer', context: 'TrainContext'):
-        if context.current_iter != self.last_ckpt_iter:
+        if context.batch_iter != self.last_ckpt_iter:
             self._save_checkpoint(trainer, context)
 
 
@@ -177,7 +179,7 @@ class StepMonitorCallback(TrainCallback):
         log_data = {
             "timestamp": time.strftime('%Y-%m-%d %H:%M:%S'),
             "epoch": context.epoch,
-            "iter": context.current_iter,
+            "iter": context.batch_iter,
             "metrics": self.metrics,
         }
         
@@ -207,7 +209,7 @@ class StepMonitorCallback(TrainCallback):
         """ Logs training information to console and file. """
         log_data = self._handle_info(trainer, context)
         try:
-            log_file = self.log_dir / f"log_epoch_{context.epoch}_iter_{context.current_iter}.json"
+            log_file = self.log_dir / f"log_epoch_{context.epoch}_iter_{context.batch_iter}.json"
             with open(log_file, 'a') as f:
                 json.dump(log_data, f, indent=4)
         except Exception:
@@ -218,4 +220,3 @@ class StepMonitorCallback(TrainCallback):
             self._handle_log(trainer, context)
         
         self.step_num += 1
-        

khaosz/trainer/train_context.py

@@ -17,7 +17,7 @@ class TrainContext:
     scheduler: BaseScheduler = field(default=None)
     checkpoint: Checkpoint = field(default=None)
     epoch: int = field(default=0)
-    current_iter: int = field(default=0)
+    batch_iter: int = field(default=0)
     loss: float = field(default=0.0)
     
     def asdict(self) -> dict:
@@ -37,6 +37,10 @@ class TrainContextBuilder:
                 tokenizer=self.trainer.parameter.tokenizer,
                 config=self.trainer.parameter.config,
             )
+        else:
+            self._context.epoch = checkpoint.epoch
+            self._context.batch_iter = checkpoint.batch_iter
+        
         self._context.checkpoint = checkpoint
         return self
     
@@ -70,10 +74,12 @@ class TrainContextBuilder:
         return self
     
     def with_dataloader(self) -> Self:
+        # fix: change batch level batch_iter to sample level offset
+        sampler_offset = self._context.batch_iter * self.trainer.train_config.batch_size
         resumeable_sampler = ResumeableRandomSampler(
             data_source=self.trainer.train_config.dataset,
             start_epoch=self._context.epoch,
-            start_iter=self._context.current_iter,
+            start_iter=sampler_offset,
             seed=self.trainer.train_config.random_seed
         )
         

khaosz/trainer/trainer.py

@@ -65,7 +65,7 @@ class Trainer:
                 self._call_callbacks('on_epoch_begin', context)
                 
                 for batch in context.dataloader:
-                    if context.current_iter % self.train_config.accumulation_steps == 0:
+                    if context.batch_iter % self.train_config.accumulation_steps == 0:
                         # 2. step
                         self._call_callbacks('on_step_begin', context)
                         self.train_config.optimizer.step()
@@ -76,7 +76,7 @@ class Trainer:
                     self._call_callbacks('on_batch_begin', context)
                     loss = self.train_config.strategy(batch)
                     context.loss = loss.item()
-                    context.current_iter += 1
+                    context.batch_iter += 1
                     
                     # to make the loss normalized by accumulation steps
                     normalized_loss = loss / self.train_config.accumulation_steps